Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74997.86402027388,
-      "required_pages_per_second_upper_estimate": 0.005188526251942216
+      "required_token_positions_per_second": 77022.5547403701,
+      "required_pages_per_second_upper_estimate": 0.005328599053888192
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9374.733002534234,
-      "required_pages_per_second_upper_estimate": 0.003925004941474619
+      "required_token_positions_per_second": 9627.819342546263,
+      "required_pages_per_second_upper_estimate": 0.004030966906993894
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18749.46600506847,
-      "required_pages_per_second_upper_estimate": 0.0039050193121864378
+      "required_token_positions_per_second": 19255.638685092526,
+      "required_pages_per_second_upper_estimate": 0.004010441732764217
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28124.199007602703,
-      "required_pages_per_second_upper_estimate": 0.0016860027873403874
+      "required_token_positions_per_second": 28883.45802763879,
+      "required_pages_per_second_upper_estimate": 0.0017315192062701541
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28124.199007602703,
-      "required_pages_per_second_upper_estimate": 0.0021865327675632527
+      "required_token_positions_per_second": 28883.45802763879,
+      "required_pages_per_second_upper_estimate": 0.0022455618167435717
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28124.199007602703,
-      "required_pages_per_second_upper_estimate": 0.3607470274573531
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28883.45802763879,
+      "required_pages_per_second_upper_estimate": 0.3704859869375558
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08537972302343544,
-    "producer_p95_seconds": 0.12155711299999439,
-    "gpu_update_seconds": 0.9830679975108277,
-    "producer_batches_per_second": 11.712382806928469,
-    "gpu_batches_per_second": 1.0172236330874822,
-    "producer_headroom": 11.514068712087418,
+    "producer_batch_seconds": 0.3588128503671877,
+    "producer_p95_seconds": 0.12907270899995638,
+    "gpu_update_seconds": 0.9572261040746378,
+    "producer_batches_per_second": 2.7869681896193508,
+    "gpu_batches_per_second": 1.044685258522815,
+    "producer_headroom": 2.6677587023292775,
     "sustainable": true,
-    "buffer_seconds": 60.95021584567132,
+    "buffer_seconds": 60.305244556702185,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 78,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 46,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 757739.5161216174,
+    "generation_bytes_per_second": 242871.18008123705,
     "upload_bytes_per_second": 770242.1525414627,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.29663044326969,
+    "estimated_full_shards_per_hour": 52.11450149371942,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -1,3 +1,4 @@
 {"step": 397200, "loss": 0.6217398047447205, "lr": 0.0003, "grad_norm": 0.1379693229902149, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58139220888, "tokens_per_second": 152102.9842612424, "data_wait_s": 1.3022000075579854e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 397300, "loss": 0.5580515265464783, "lr": 0.0003, "grad_norm": 0.18260739819030208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58154046642, "tokens_per_second": 153952.47086143732, "data_wait_s": 1.124099992466654e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 397400, "loss": 0.5864430665969849, "lr": 0.0003, "grad_norm": 0.16819411572980625, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58168497235, "tokens_per_second": 148897.31773298592, "data_wait_s": 1.0385999985373928e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 397500, "loss": 0.5782780051231384, "lr": 0.0003, "grad_norm": 0.14115527360265942, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58183021172, "tokens_per_second": 149143.60326988477, "data_wait_s": 1.2393000019983447e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:108938a048ad14cf7f9a45aa4fd0c4e97ad363c61663c9fdcbc6974d0b66532e
-size 6519
+oid sha256:202e38262e34c7ffa2bf7bb7f0714346ea3f7cbb7842f8cdef714c4a1c29b8f9
+size 8637